WO2018100655A1

WO2018100655A1 - Data collection system, abnormality detection system, and gateway device

Info

Publication number: WO2018100655A1
Application number: PCT/JP2016/085480
Authority: WO
Inventors: 緒方　祐次; 大介石井
Original assignee: 株式会社日立製作所
Priority date: 2016-11-30
Filing date: 2016-11-30
Publication date: 2018-06-07
Also published as: US20190204818A1; US11067973B2; JP6675014B2; JPWO2018100655A1

Abstract

A data collection system according to one embodiment of the present invention collects time-sequential data outputted from a sensor that is provided to a facility to be monitored, and carries out facility abnormality detection. The data collection system has stored therein a plurality of templates which are data sets to be compared with the time-sequential data, and determines the inspection range of the time-sequential data by comparing the time-sequential data with the plurality of templates in a learning step. In an abnormality detection step, a frequency spectrum to be inspected is extracted from among frequency spectra of the time-sequential data by using information about the inspection range of the time-sequential data determined in the learning step, and the facility abnormality detection is carried out by using the extracted frequency spectrum.

Description

Data collection system, abnormality detection method, and gateway device

The present invention relates to a technique for collecting and processing data obtained from a sensor.

In plants and industrial facilities, a technology is used in which a large number of sensors are installed in machinery and equipment, and the sensor data is collected by a computer and the computer diagnoses the machinery and equipment.

For example, Patent Document 1 discloses an abnormality diagnosis system that diagnoses an abnormality of a bearing or a bearing-related member in a mechanical facility by detecting sound or vibration generated from the mechanical facility with a sensor and analyzing the detection signal. Yes. The abnormality diagnosis system includes an envelope processing unit that obtains an envelope of a detection signal, an FFT unit that converts the envelope obtained by the envelope processing unit into a frequency spectrum, and a moving average process for the frequency spectrum obtained by the FFT unit. A peak detecting unit that smoothes and detects the peak, and a diagnostic unit that diagnoses an abnormality based on the peak of the frequency spectrum detected by the peak detecting unit.

JP 2006-111302 A

In the machine equipment diagnosis system, the sensor data of each device becomes enormous, and in the system that detects anomalies from these enormous data, there is a problem that the processing cost and the storage cost increase. In the technique disclosed in Patent Document 1, since data in the entire frequency range is collected and processed, it is difficult to keep data processing costs low.

The data collection system according to an embodiment of the present invention collects time-series data output from a sensor provided in a facility to be monitored and detects an abnormality of the facility. The data collection system stores multiple models that are data for comparison with time-series data, and determines the inspection range of time-series data by comparing the time-series data with multiple models in the learning process. To do. In the abnormality detection step, the information on the inspection range of the time series data determined in the learning step is used to extract the frequency spectrum to be inspected from the frequency spectrum of the time series data, and the extracted frequency spectrum is used to extract the frequency spectrum. Detect equipment abnormalities.

The data collection system according to an embodiment of the present invention detects an abnormality of equipment using only a part of the information in the frequency spectrum of the time series data collected from the sensor, thereby increasing the processing cost. Can be suppressed.

It is a block diagram which shows the outline | summary of a data collection system. It is a block diagram which shows the function structural example of a data collection system. It is a block diagram which shows an example of the hardware constitutions of a center facility server. It is a block diagram which shows an example of the hardware constitutions of a gateway. It is a flowchart which shows an example of the process performed in the learning information change part of a central facility server. It is a figure which shows an example of a sensor data table. It is a figure which shows an example of a frequency analysis data table. It is a figure which shows an example of a frequency analysis parameter table. It is a figure which shows an example of a similarity management table. It is a figure which shows an example of the learning frequency management table. It is a figure which shows an example of the waveform of a model. It is a figure which shows an example of the waveform of a model. It is a figure which shows an example of the waveform of sensor data. It is a figure which shows an example of a learning information management table. It is a flowchart which shows an example of the process performed in the learning information selection part of a gateway. It is a figure which shows an example of a GW learning information management table. It is a figure which shows the example of the GW learning information management table updated by the learning information selection part. It is a figure which shows the example of the GW learning information management table updated by the learning information selection part. It is a figure which shows an example of the waveform of the sensor data stored in a sensor data storage part. It is a figure which shows an example of the flow of the process performed with a central facility server and a gateway.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

FIG. 1 is a block diagram showing an outline of a data collection system according to an embodiment of the present invention. In the data collection system, a plurality of sensors 3-1 to 3-n that measure physical quantities such as vibration and temperature are provided in machines, facilities, and structures to be monitored. The data measured by the sensors 3-1 to 3-n may be abbreviated as the central facility server 1 (“server 1”) from the gateways 2-1 to 2-n via the network (Wide Area Network (WAN)) 70. Is sent). In addition, when the whole of the sensors 3-1 to 3-n is indicated, a symbol “3” without “−” is used, and when an individual sensor is specified, a symbol with a subscript “−” or less is used. The same applies to the other components.

In FIG. 1, a sensor node (wireless communication unit) 4-1 connected to the sensor 3-1 is connected to the gateway 2-1 via a FAN (Field Area Network) 80-1 which is a wireless communication network. The network connecting the sensor 3 and the sensor node 4 is not limited to wireless communication, and both may be connected via a wired communication network. The gateway 2 is a relay device that relays communication between different networks, and can be configured by a router, for example. The FAN 80 is not limited to wireless communication, and may be a network including a wired communication network. The number of sensors 3 connected to the sensor node 4 and the number of sensor nodes 4 connected to the gateway 2 are not limited to the number shown in FIG. 1 and may take various forms. Can do. For example, a plurality of sensor nodes 4 may be connected to one gateway 2. Alternatively, a plurality of sensors 3 may be connected to the sensor node 4.

In addition, the correspondence relationship between the gateway 2 and the sensor node 4 is set in advance, and the gateway 2 communicates with the subordinate sensor node 4.

One or more sensors 3 are arranged for one monitoring target, and at least one gateway 2 is installed for each monitoring target. The gateway 2 transfers time series data (hereinafter referred to as sensor data) collected from the sensor 3 to the central facility server 1 via the WAN 70. A plurality of monitoring targets may exist under the gateway 2.

Gateway 2 detects sensor abnormality by analyzing sensor data in addition to collecting sensor data. The gateway 2 monitors the state of the subordinate sensor 3 and controls the sensor 3. In the example of FIG. 1, the example in which the sensor 3 is connected via the sensor node 4 under the gateway 2 is shown, but the present invention is not limited to this. For example, when the number of sensors 3 is large, or when devices or facilities within the monitoring target are separated, a configuration in which a plurality of gateways 2 are connected in a hierarchy may be employed.

Based on the sensor data received from the gateway 2, the central facility server 1 performs a learning process for identifying a frequency range (sensing range) when the gateway 2 detects an abnormality to be monitored, and notifies the gateway 2 of the result. To do. In addition, the central facility server 1 performs information processing with added value such as monitoring of a monitoring target, visualization of sensor data, analysis of sensor data or prediction of failure of the monitoring target, and a user who maintains the monitoring target. Provide information to customers. However, in the following description of the present embodiment, the description of the monitoring target monitoring and sensor data visualization performed in the central facility server 1 will be omitted. In the following description, the processing related to the facility abnormality detection method, which is a characteristic function of the data collection system according to the present embodiment, will be mainly described.

In addition, as a monitoring object in the data collection system according to the present embodiment, in addition to monitoring a machine such as a plant, an industrial facility, a transportation device, a vending machine, a building such as a bridge, a road, and a tunnel is a monitoring object. be able to. In addition, the monitoring target of the data collection system is not limited to machines and buildings, but images and urban environment (town information) can also be monitored.

FIG. 2 is a diagram showing an example of functional blocks constituting the data collection system. In the example of FIG. 2, the central facility server 1 is connected to the gateway 2 via the WAN 70, and one or more (for example, n) sensors 5 (sensors 5-1 to 5-n) installed in the machine 9 to be monitored. ) To collect sensor data. In the following description, the set of sensor 3 and sensor node 4 described in FIG. 1 is referred to as “sensor 5”. In the following, the names of n sensors (sensor names) are “sensor # 1”, “sensor # 2”,. . . Sometimes referred to as “sensor #n”.

The sensor 5 transmits sensor data obtained by measuring the state of the machine 9 to the gateway 2. The electric power of the sensor 5 can be supplied from a battery (or storage battery) not shown. In addition, as a structure which has a solar cell panel in the sensor 5, the structure which supplies electric power to a sensor 5 from a solar cell panel may be sufficient, and the sensor 5 is not limited to the drive by a battery.

Further, the sensor 5 may be connected to the gateway 2 via a wired network or may be connected via a wireless network.

Also, the type of sensor 5 used for collecting sensor data is not limited to a specific one. An appropriate type of sensor 5 may be used according to the type of information (physical quantity or the like) that is desired to be acquired from the monitored machine 9. In the present embodiment, an example in which the data collection system is intended to detect an abnormality (or a sign of abnormality) by observing vibrations generated in the machine 9 will be described. Therefore, an example in which a sensor (for example, an acceleration sensor or the like) that can measure displacement (or information such as acceleration, which can calculate displacement) can be used will be described.

The sensor 5 measures the displacement periodically (according to the sampling period), and the measured value (displacement) is continuously transmitted to the gateway 2. That is, the sensor 5 outputs time series data of measured values. The gateway 2 adds the time (the time when each displacement was measured) to each measurement value received from the sensor 5 and outputs it to the central facility server 1. Instead of adding the time to the measurement value (displacement) received by the gateway 2 from the sensor 5, the sensor 5 may create information with the time added to the measurement value and transmit it to the gateway 2.

The gateway 2 transmits sensor data to the central facility server 1 and detects an abnormality of the monitored machine 9. Specifically, the gateway 2 performs frequency analysis of sensor data acquired from each sensor 5 to obtain a frequency spectrum, and extracts data in a predetermined frequency range from the frequency spectrum. The frequency range used at the time of extraction is initially determined by the central facility server 1 through a learning process, but may be changed by the gateway 2 thereafter. Details of the method of determining the frequency range will be described later.

Further, the gateway 2 specifies the frequency (hereinafter referred to as “frequency peak”) having the largest amplitude (or vibration intensity) from the extracted data. The gateway 2 calculates the shift speed of the frequency peak by repeatedly performing the process of specifying the frequency peak. The gateway 2 determines that an abnormality has occurred when the shift speed exceeds a preset value, and notifies the central facility server 1 that an abnormality has occurred.

The gateway 2 is connected to an operation management terminal 63 including an input / output device such as a keyboard and a display in order to check information of the subordinate sensors 5 and rewrite setting information. The operation management terminal 63 is used by an on-site worker or the like. An on-site worker operates the gateway 2 using the input / output device of the operation management terminal 63.

The functional elements of Gateway 2 are described below. The gateway 2 includes a sensor receiving unit 220 that receives sensor data from the sensor 5, a sensor data analyzing process, specifically, a sensor data analyzing unit 230 that performs frequency analysis by fast Fourier transform (FFT), and the WAN 70. A data transmission unit 240 that transmits the received sensor data to the central facility server 1, a learning information reception unit 250 that receives learning information from the central facility server 1 via the WAN 70, and a learning information selection unit that selects an analysis range of the sensor data 260, a sensor data storage unit 270 that temporarily stores sensor data, and a learning information management unit 280 that stores learning information. These functional elements are implemented by software (program).

When the gateway 2 performs polling to collect sensor data from the sensor 5, the sensor receiver 220 collects sensor data from the sensor 5 based on a predetermined sampling period of sensor data. Note that the sampling period may be different for each sensor 5.

Although the case where the gateway 2 performs the polling of the sensor data from the sensor 5 has been described, the gateway 2 transmits the sampling frequency (or sampling period) of the sensor data to the sensor 5, and the sensor 5 receives the sensor data. The configuration may be such that sensing is performed based on the sampling frequency (or sampling period).

The functional elements of the central facility server 1 that collects sensor data from the gateway 2 via the WAN 70 will be described below.

The central facility server 1 monitors the machine 9 based on the sensor data received from the gateway 2 and outputs information such as monitoring results to the central monitoring terminal 64. The central monitoring terminal 64 includes input / output devices such as a keyboard and a display.

The central facility server 1 receives the sensor data transmitted from the gateway 2 and accumulates the sensor data, and the FFT module 170 (hereinafter, the frequency analysis of the sensor data of the sensor 5 based on the received sensor data). Abbreviated as “FFT170”) and a frequency waveform section 180 for accumulating the result of frequency analysis. Similar to the sensor data analysis unit 230, the FFT 170 performs frequency analysis by fast Fourier transform.

Furthermore, the central facility server 1 includes a learning information storage unit 130 that stores information used for learning processing, a learning information change unit 120 that creates (or changes) learning information, and a learning information transmission unit that transmits learning information to the gateway 2. 190. Each table included in the learning information storage unit 130 will be described later. These functional elements are implemented by software (program).

In addition, a central monitoring and maintenance terminal 62 is connected to the central facility server 1. The central monitoring and maintenance terminal 62 is a terminal for maintenance personnel to write and update information in the learning information storage unit 130, and includes an input / output device such as a keyboard and a display, like the central monitoring terminal 64.

Note that the numbers of the central facility server 1, the gateway 2, the sensors 5, and the machines 9 are not limited to the numbers shown in FIG. Further, the connection locations of the central monitoring and maintenance terminal 62, the central monitoring terminal 64, and the operation management terminal 63 are not limited to the positions shown in FIG.

FIG. 3 is a block diagram showing an example of the configuration of the central facility server 1. The central facility server 1 includes a processor (CPU) 11 that performs arithmetic processing, a memory 12 that stores programs and data, an I / O interface 13 that is connected to the CPU 11, and programs and programs that are connected to the I / O interface 13. A storage device 14 that holds data; and a communication device 15 that is connected to the I / O interface 13 and communicates with the WAN 70. In this embodiment, an apparatus connected to the CPU 11 via the I / O interface 13 such as the storage apparatus 14 and the communication apparatus 15 is referred to as an “I / O device”.

The I / O interface 13 is composed of, for example, a controller device according to the PCI express standard, and performs communication between the CPU 11 and the I / O device.

The memory 12 is loaded with the OS 310 and the learning information change program 300 and executed by the CPU 11. Specifically, the OS 310 and the learning information change program 300 are stored in the storage device 14, loaded into the memory 12 when the central facility server 1 is activated, and executed by the CPU 11. The central monitoring / maintenance terminal 62 and the central monitoring terminal 64 shown in FIG. 2 are connected to the central facility server 1 via a LAN (not shown). Alternatively, the central monitoring and maintenance terminal 62 and the central monitoring terminal 64 may be connected to the central facility server 1 via the communication device 15.

As described above, the functional elements of the sensor receiving unit 110, the FFT 170, the frequency waveform unit 180, the learning information storage unit 130, the learning information changing unit 120, and the learning information transmitting unit 190 shown in FIG. 2 are software (programs). Implemented as That is, the central facility server 1 functions as a device having each functional element shown in FIG. 2 when the CPU 11 executes the learning information change program 300 using the memory 12 or the I / O device.

In the present specification, the process may be described with a functional element such as the learning information changing unit 120 as an operation subject. However, as described above, each functional element is a function realized by the program (learning information change program 300) being executed by the CPU 11, and therefore, the functional element in the central facility server 1 is described as an operation subject. This means that the process is actually executed by the CPU 11. The learning information change program 300 may be provided in a state of being stored in a computer-readable non-transitory data storage medium such as an IC card, an SD card, or a DVD.

FIG. 4 is a block diagram showing an example of the configuration of the gateway 2. The gateway 2 communicates with the CPU 21 that performs arithmetic processing, the memory 22 that stores programs and data, the I / O interface 23 connected to the CPU 21, and the WAN 70 connected to the I / O interface 23. A WAN-side communication device 24, a sensor-side communication device 25 that is connected to the I / O interface 23 and communicates with the sensor 5, and a storage device 26 that is connected to the I / O interface 23 and holds programs and data. And including. The I / O interface 23 is composed of, for example, a controller device according to the PCI express standard, and performs communication between the CPU 21 and the I / O device. The operation management terminal 63 shown in FIG. 2 is connected to the gateway 2 via a LAN (not shown).

The OS 290 and the learning information selection program 400 are loaded into the memory 22 and executed by the CPU 21.

As with the central facility server 1, each functional element of the gateway 2 is implemented as software. That is, the CPU 21 of the gateway 2 executes the learning information selection program 400 while using the memory 22 and the I / O device, so that the gateway 2 is connected to the sensor reception unit 220, the sensor data analysis unit 230, and the data shown in FIG. The transmission unit 240, the learning information reception unit 250, the learning information selection unit 260, and the sensor data storage unit 270 are caused to function as an apparatus including the functional elements. For this reason, in this specification, the processing may be described with the functional elements in the gateway 2 such as the learning information selection unit 260 as the operating subject, but the processing described with the functional elements in the gateway 2 as the operating subject is as follows. In practice, this means that the processing is performed by the CPU 21. The learning information selection program 400 may be provided in a state of being stored in a computer-readable non-transitory data storage medium such as an IC card, an SD card, or a DVD.

FIG. 5 is a flowchart illustrating an example of processing performed by the learning information changing unit 120 of the central facility server 1. This process is executed when the central facility server 1 receives a predetermined amount of sensor data from the gateway 2 via the WAN 70 (specifically, an amount equal to or greater than the FFT length). This process is executed for each sensor.

Before explaining the flow shown in FIG. 5, some information used in the learning information changing unit 120 will be explained. Sensor data (time-series data of measurement values) transmitted from the gateway 2 is stored in a sensor data table 600 included in the sensor reception unit 110. An example of the sensor data table 600 is shown in FIG. Each row (record) of the sensor data table 600 includes a time 601, a sensor # 1 (602), a sensor # 2 (603),. . . It has a column for sensor #n (604). Sensor # 1 (602), Sensor # 2 (603),. . . In sensor #n (604), the measurement value (displacement) by each sensor 5 is registered, and time 601 represents the time when the measurement values stored in columns 602 to 604 were measured.

The learning information changing unit 120 performs frequency analysis (FFT) of sensor data received from the gateway 2 for a predetermined period (referred to as “learning time”), and stores the frequency-analyzed data and the data stored in advance. Compare with the template which is the comparison information. First, the learning number management table 1000 in which information related to learning time is recorded will be described.

FIG. 10 shows an example of the learning count management table 1000. The learning number management table 1000 is a table that the learning information storage unit 130 has, and is a table for managing learning time information for each sensor. A sensor name is stored in the sensor 1001, and a learning time using sensor data obtained from the sensor specified by the sensor 1001 is stored in the learning time 1002. Information stored in the sensor 1001 and the learning time 1002 is designated by the worker. Specifically, the operator uses the central monitoring and maintenance terminal 62 to set a sensor name and a learning time for the sensor 1001 and the learning time 1002 in the learning number management table 1000.

On the other hand, the learning start time 1003 and the learning end time 1004 are set when the learning information changing unit 120 starts the process of FIG. The contents of information set in the learning start time 1003 and the learning end time 1004 will be described later.

Next, the frequency analysis parameter table 800 will be described. The frequency analysis parameter table 800 is a table that stores information necessary for frequency analysis, such as a sampling frequency, for each sensor 5, and is a table that the learning information storage unit 130 has.

FIG. 8 shows an example of the frequency analysis parameter table 800. Each row of the frequency analysis parameter table 800 includes a column of a sensor 801, a sampling frequency 802, a sampling period 803, and an FFT length 804. These pieces of information are information set by the operator using the central monitoring and maintenance terminal 62. These pieces of information are information set for each sensor 5.

Sampling frequency 802, sampling period 803, and FFT length 804 are information used in frequency analysis (FFT) (that is, sampling frequency, sampling period, FFT length), and sensor 801 stores the sensor name. For example, when the learning information changing unit 120 performs frequency analysis of sensor data collected from “sensor # 1”, the sampling frequency 802 (or sampling period 803) of the row in which “sensor # 1” is stored in the sensor 801, Frequency analysis is performed according to the FFT length 804.

In addition to this, there is information used by the learning information changing unit 120, which will be described in the process of explaining the processing flow of the learning information changing unit 120.

From here, the process performed by the learning information changing unit 120 will be described. In the following description, unless otherwise specified, the learning information changing unit 120 processes the sensor data collected by a certain sensor 5 (assuming the sensor name of the sensor 5 is “sensor #s”). An example of performing this will be described. When execution of the learning information changing unit 120 is started, first, a value is registered in the learning number management table 1000 included in the learning information storage unit 130 (step 500). In step 500, the learning information changing unit 120 registers the current time in the learning start time 1003, and the learning end time 1004 is a value obtained by adding the learning time indicated by the learning time 1002 to the time registered in the learning start time 1003 ( Time).

Subsequently, the learning information changing unit 120 receives the sensor data of the sensor #s from the sensor receiving unit 110, performs frequency analysis of the sensor data using the FFT 170, and obtains a result of the frequency analysis (frequency spectrum) in the frequency waveform unit 180. accumulate. (Step 501). In the frequency analysis, the learning information changing unit 120 (and the FFT 170) performs analysis according to information stored in the frequency analysis parameter table 800.

FIG. 13 shows an example of the result (frequency spectrum) of the frequency analysis of the sensor data stored in the frequency waveform section 180. FIG. 13 is a graphical representation of a frequency spectrum, where the vertical axis represents vibration intensity (signal intensity) and the horizontal axis represents frequency. In this embodiment, the frequency spectrum of the sensor data or a graph representation of this frequency spectrum is called a “waveform”. In the following description, the waveform obtained by frequency analysis of the sensor data of sensor #s obtained in step 501 will be referred to as “sensor #s waveform”.

In FIG. 13, the frequency analysis result of the sensor data is represented in a graph for the sake of explanation. However, information (graphics, images, etc.) represented in a graph as shown in FIG. 13 is not necessarily recorded in the frequency waveform unit 180. There is no need. The frequency waveform section 180 may record a data series of frequency and vibration intensity.

11 and 12 are examples of template waveforms. In the data collection system according to the present embodiment, for example, sensor data collected from various machines in the past until the failure of the machine is held in advance, and in the present embodiment, this sensor data is referred to as a “model”. . The template is used for comparison with the frequency spectrum of the sensor data obtained in step 501. The waveform of the template means a frequency spectrum (or a graph representation thereof) obtained by frequency analysis of sensor data collected from various machines in the past. 1101 to 1103 in FIG. 11 and 1200 in FIG. 12 are graph representations of the frequency spectrum obtained by frequency analysis of the template, as described with reference to FIG.

The model is stored in the learning information storage unit 130 (more precisely, it is stored in the frequency analysis data table 700 of the learning information storage unit 130). Any model data storage format may be adopted. For example, the result (frequency spectrum) obtained by frequency analysis of the model may be stored in the learning information storage unit 130, or the model itself (data before frequency analysis) may be stored in the learning information storage unit 130. Further, the template does not necessarily have to be sensor data collected from an actual machine. For example, artificially created data may be used as a template by simulation or the like.

In the following description, an example will be described in which a template is sensor data collected from an actual machine, and the learning information storage unit 130 stores a result of frequency analysis of the template (model waveform). However, in order to avoid redundant description, the learning information storage unit 130 may be expressed as “model” being stored. *

In the data collection system according to the present embodiment, a plurality of types (for example, n types) of templates are stored in the learning information storage unit 130. For example, sensor data collected from a plurality of types of machines is stored as a template. In the following, a plurality of templates are respectively referred to as “model 1”, “model 2”,. . . “Model n”, “Model 1”, “Model 2”,. . . Etc. are also called template identifiers.

Also, the learning information storage unit 130 stores frequency analysis results at a plurality of points for each template. In the present embodiment, a template waveform at a certain time (referred to as tf0) and a time (referred to as tfm) after elapse of m hours (or m seconds or m minutes) from time tf0, where m is a predetermined natural number. A description will be given of an example in which a plurality of template waveforms obtained up to this point are stored. Specifically, times tf0, tf1, tf2,. . . The waveform of the template at tfm is stored in the learning information storage unit 130 for each template. Note that the acquisition time of each template need not be the same time. As long as the waveform of each template is stored in the learning information storage unit 130 for a period of m hours (or m seconds, m minutes) from a certain point in time, the acquisition time of the sensor data of each template is different. Also good. Note that reference numerals 1101 to 1103 in FIG. 11 represent the waveforms of template 1, template 2, and template n at time tf0, respectively, while 1200 in FIG. 12 represents the waveform of template 1 at time tfm.

The learning information storage unit 130 has a frequency analysis data table 700 for storing templates. FIG. 7 shows an example of the frequency analysis data table 700. Each row (record) of the frequency analysis data table 700 stores a frequency spectrum of a template at a certain point in time. The data series 705 in each row stores a plurality of sets of frequencies and their vibration intensities as the frequency spectrum of the time model stored at the time 701. The frequency analysis data table 700 is provided for each template. When there are n templates, the learning information storage unit 130 has n frequency analysis data tables 700.

Further, a median value 702, a lower limit value 703, and an upper limit value 704 are stored in each row of the frequency analysis data table 700. The definitions of the median value, the lower limit value, and the upper limit value in the present example are as follows. In the waveform (frequency spectrum obtained by frequency analysis of sensor data), the frequency at the point where the vibration intensity is maximum is called the “median value”. Also, in the waveform, the frequency at the lowest frequency among the points where the vibration intensity is minimum is called the “lower limit value”, and the frequency at the highest frequency among the points where the vibration intensity is minimum is It is called “upper limit value”. The lower limit value and the upper limit value are information used for narrowing down information used when the gateway 2 detects abnormality of the machine 9 using the sensor data. Details will be described later.

In this embodiment, the lower limit value, the upper limit value, and the median value of the waveform at time tfk (k is an integer of 0 or more) are expressed as “f1_k”, “f2_k”, and “fc_k”, respectively. The graph shown in FIG. 11 shows an example of the waveform at time tf0 of

templates

1, 2, and n. The points at frequencies fc_0, f1_0, and f2_0 are the median, lower limit, and upper limit, respectively. is there.

In step 502, the learning information changing unit 120 compares the sensor data result stored in the frequency waveform unit 180 with the template stored in the frequency analysis data table 700, and calculates the correct answer rate. The correct answer rate is calculated for each template.

The correct answer rate obtained in step 502 is registered in the similarity management table 900 shown in FIG. The similarity management table 900 is also one of the tables included in the learning information storage unit 130, and the similarity management table 900 is provided for each sensor. Each row of the similarity management table 900 includes columns of a template 901, a correct answer rate 903, and a similarity candidate 904. The template 901 stores a template identifier.

In calculating the correct answer rate, in step 502, the waveform of the sensor #s is compared with the waveform of each template, and the correct answer rate calculated from the comparison result is registered in the correct answer rate 903. The correct answer rate may be obtained by a predetermined calculation formula. Note that the correct answer rate (value recorded in the correct answer rate 903) calculated by the learning information changing unit 120 in the present embodiment is, for example, a value between 0 and 1. The closer the correct answer rate is to 1, the closer the frequency analysis result of the sensor data is to the template waveform.

As described above, the learning information storage unit 130 stores the waveform of the template at each time (the waveform of the template at times tf0, tf1, tf2,... Tfm) for each template. When the learning information changing unit 120 compares the waveform of the template a (1 ≦ a ≦ n), for example, in step 502, the waveform of the template a at each time is compared with the waveform accumulated in the frequency waveform unit 180. Thus, a plurality of similarities (similarity is a value of 0 or more and 1 or less, for example, the closer the value is to 1, the more similar the waveforms are), and the maximum value among them is determined for the waveform of the template a. The correct answer rate is determined. Since the learning information changing unit 120 calculates the correct answer rate for all templates (

models

1, 2,... N), n correct answer rates are finally obtained in step 502. In addition, as a method for determining the similarity between a data series such as the analysis result of sensor data accumulated in the frequency waveform unit 180 and another data string, various statistical methods such as a dynamic time warping method are known, Here, any of those known methods may be used.

In step 503, if the correct answer rate obtained in step 502 is greater than or equal to a preset correct answer rate threshold for each template, the learning information changing unit 120 registers “o” in the similar candidate 904 and is set in advance. If the correct answer rate is less than the threshold, “x” is registered in the similar candidate 904.

In step 504, the learning information changing unit 120 determines whether or not the current time (the time when the step 504 is executed) has reached the learning end time 1004. If not, the process proceeds to step 501. Transitions to step 505. The determination as to whether or not the learning end time implemented in step 504 has been reached is made based on the information in the learning count management table 1000 shown in FIG.

In step 505, the learning information changing unit 120 determines that the waveform acquired from the sensor 5 is similar to any waveform of the template based on the correct answer rate 903 and the similarity candidate 904 information in the similarity management table 900 determined in step 503. And determine a similar template. When “◯” is registered in a plurality of similar candidates 904, a template having the highest correct answer rate 903 is selected as a similar template.

In step 506, the learning information changing unit 120 reads the median value, the lower limit value, and the upper limit value at the time tf0 of the similar template determined in step 505 from the frequency analysis data table 700.

In step 507, the learning information changing unit 120 reads the median value, lower limit value, and upper limit value at the time tfm of the similar template determined in step 505 from the frequency analysis data table 700.

For example, when the waveform 1101 (FIG. 11) at time tf0 of the template 1 and the waveform 1300 of the sensor # 1 of FIG. 13 are determined to be similar in step 505, the learning information changing unit 120 in step 506 The median value 702 (fc_0), the lower limit value 703 (f1_0), and the upper limit value 704 (f2_0) of the row where the time 701 is “tf0” are read from the frequency analysis data table 700. Further, in step 507, the frequency analysis data of the template 1 is read. From the table 700, the median value 702 (fc_m), the lower limit value 703 (f1_m), and the upper limit value 704 (f2_m) of the row where the time 701 is “tfm” are read out.

In step 508, the learning information changing unit 120 calculates the shift speed of the frequency peak value (median value) using the two median values (fc_0, fc_m) of the template obtained in steps 506 and 507. The shift speed is a value representing the change amount of the median value per unit time, and is obtained by calculating (fc_m−fc_0) / (tfm−tf0). As another embodiment, the central facility server 1 may hold the shift speed of the median value in the frequency analysis data table 700 in advance. In that case, the learning information changing unit 120 does not need to calculate the shift speed in step 508, and may read the shift speed from the frequency analysis data table 700.

In step 509, the learning information changing unit 120 reads the median value, the lower limit value, and the upper limit value read in step 506, the shift speed calculated in step 508, and each sensor 801 registered in advance in the frequency analysis parameter table 800. Are stored in the learning information management table 1400 and transmitted to the gateway 2.

The contents of the learning information management table 1400 will be described with reference to FIG. The learning information management table 1400 is a table for storing information such as the median value and the shift speed obtained up to step 508 for each sensor 5, and is also a table included in the learning information accumulation unit 130. The learning information management table 1400 includes columns of a sensor 1401, a learning operation state 1402, a median value 1403, a lower limit value 1404, an upper limit value 1405, a shift speed 1406, a sampling frequency 1407, a sampling period 1408, and an FFT length 1409.

In each of the sensor 1401, the sampling frequency 1407, the sampling period 1408, and the FFT length 1409, the same information as the information registered in the sensor 801, the sampling frequency 802, the sampling period 803, and the FFT length 804 of the frequency analysis parameter table 800 is stored. Has been. In the learning operation state 1402, “non-operation” is initially stored. The median value 1403, the lower limit value 1404, the upper limit value 1405, and the shift speed 1406 are columns for storing the median value, lower limit value, upper limit value, and shift speed of similar templates, which are obtained in steps 505 to 508 described above. No value is stored until step 509 is executed (in FIG. 14 (or FIG. 16 to FIG. 18 described later), the column storing “none” indicates that no value is stored). Means).

In step 509, the learning information changing unit 120 stores information in each field of the row “sensor #s” by the sensor 1401 among the rows in the learning information management table 1400. Specifically, the median value 1403, the lower limit value 1404, and the upper limit value 1405 store the median value, lower limit value, and upper limit value of the similar template read in step 506, and the shift speed 1406 is determined in step 508. In addition, the learning operation state 1402 stores “operation”. Further, in the learning information change unit 120, the sensor 1401 of the learning information management table 1400 transmits information (sensor 1401 to FFT length 1409) of each field in the row “sensor #s” to the gateway 2. In this embodiment, the information transmitted here is called “learning information”. Up to this point, the learning information creation and transmission process by the learning information changing unit 120 ends.

In step 510, the learning information changing unit 120 receives the sensor data from the gateway 2 and accumulates it in the sensor receiving unit 110. Note that the gateway 2 may or may not transmit sensor data to the central facility server 1. This depends on the setting of the gateway 2 (details will be described later). If the gateway 2 is set not to transmit the sensor data to the central facility server 1, the process of step 510 is not performed.

In step 511, the learning information changing unit 120 determines whether or not to shift to the learning mode. The learning mode is a mode in which the learning information changing unit 120 executes the processing from step 500 to step 509 described above, and the instruction to shift to the learning mode is sent to the central monitoring and maintenance terminal 62 of the central facility server 1. Via maintenance personnel. When the maintenance engineer instructs to shift to the learning mode (step 511: YES), the learning information changing unit 120 performs the processing from step 500 again. If the maintenance staff has not instructed the shift to the learning mode (step 511: NO), step 510 is executed again.

In the present embodiment, the learning information changing unit 120 is provided in the central facility server 1 and the central facility server 1 performs the processing shown in FIG. 5, but the learning information changing unit 120 (FIG. 5). This process may be performed by other than the central facility server 1. For example, it may be performed in the gateway 2 or may be performed in other devices. However, when the processing performance of the gateway 2 is low, the central facility server 1 can be made to perform the processing of the learning information changing unit 120 (FIG. 5) so as not to apply an excessive load to the gateway 2. preferable.

FIG. 15 is a flowchart illustrating an example of processing performed by the learning information selection unit 260 of the gateway 2. This process is executed when sensor data is received from the sensor 5. This process is also performed for each sensor 5. Before the description of the flow illustrated in FIG. 15, first, the GW learning information management table 1600 included in the learning information management unit 280 of the gateway 2 will be described.

FIG. 16 shows an example of the GW learning information management table 1600. The GW learning information management table 1600 is a table similar to the learning information management table 1400 described above, and is a table for storing learning information transmitted from the central facility server 1 (information transmitted in step 509). . Further, the contents of the GW learning information management table 1600 are updated each time the process of FIG.

Among the columns of the GW learning information management table 1600, the sensor 1601, the learning operation state 1602, the median value 1603, the lower limit value 1604, the upper limit value 1605, the shift speed 1606, the sampling frequency 1608, the sampling period 1609, and the FFT length 1610 are learning information. The management table 1400 is a column that stores the same information as the sensor 1401, the learning operation state 1402, the median value 1403, the lower limit value 1404, the upper limit value 1405, the shift speed 1406, the sampling frequency 1407, the sampling period 1408, and the FFT length 1409. . In each of these columns, the learning information receiving unit 250 stores learning information transmitted from the central facility server 1 (that is, information on the sensors 1401 to FFT length 1409).

16 is a row in which information about sensor # 1 is stored, and row 1622 is a row in which information about sensor # 2 is stored. A row 1621 represents a state in which learning information is transmitted from the central facility server 1, and a row 1622 represents a state in which learning information has not yet been transmitted from the central facility server 1. As shown in the row 1622, before the learning information is transmitted from the central facility server 1, the columns 1603 to 1610 are in a state where no information is stored, the sensor 1601, the learning operation state 1602, and the sensor data transmission. 1611 stores information only. Each of the sensor 1601 and the learning operation state 1602 stores a sensor name and “non-operation”.

On the other hand, “Yes” or “No” is stored in the sensor data transmission 1611, and when “Yes” is stored, the gateway 2 transmits the sensor data to the central facility server 1. Information stored in the sensor data transmission 1611 may be determined, for example, by a worker on site. For example, if the sensor data of sensor # 1 is not stored in the central facility server 1 but the sensor data of sensor # 2 is stored in the central facility server 1, the worker uses the operation management terminal 63 as shown in FIG. In addition, “None” may be set in the sensor data transmission 1611 in the row 1621, and “Yes” may be set in the sensor data transmission 1611 in the row 1622.

When the learning operation state 1602 is “non-operation”, the sensor data is transmitted from the gateway 2 to the central facility server 1 even if “none” is set in the sensor data transmission 1611. This is because the central facility server 1 performs the learning process.

Also, the median value 1603, the lower limit value 1604, the upper limit value 1605, and the shift speed current value 1607 are updated each time the learning information selection unit 260 executes processing to be described later.

Hereinafter, the flow of processing performed by the learning information selection unit 260 will be described with reference to FIG. In the following description, unless otherwise specified, the learning information selection unit 260 processes the sensor data collected by a certain sensor 5 (assuming the sensor name of the sensor 5 is “sensor #s”). An example of performing this will be described.

In step 1501, the learning information selection unit 260 receives the sensor data at the sensor reception unit 220, accumulates the acquired sensor data in the sensor data storage unit 270, and stores the sensor data in the central facility server 1 via the data transmission unit 240. Send.

In step 1502, the learning information selection unit 260 determines whether or not it is in a learning operation state. Whether or not the learning operation state is set is determined based on a value registered in the learning operation state 1602 of the GW learning information management table 1600. When the learning information of the sensor #s is transmitted to the gateway 2 by executing the above-described step 509 in the central facility server 1, the gateway 2 performs the line for the sensor #s in the GW learning information management table 1600. The learning information is stored in. As a result, the learning operation state 1602 of the sensor #s becomes “operation”.

When the value registered in the learning operation state 1602 of the GW learning information management table 1600 is “operation”, the learning information selection unit 260 next executes step 1503 to enter the learning operation state 1602 of the GW learning information management table 1600. If the registered value is “non-operating”, the process returns to step 1501. That is, when not in the learning operation state (until learning information is sent from the central facility server 1), the gateway 2 performs only the process of transmitting the sensor data received from the sensor 5 to the central facility server 1.

The processing after step 1503 is performed when it is determined that the learning operation state is determined by the determination result of step 1502. Note that the processing after step 1503 is repeatedly executed unless it is determined to be abnormal in step 1509 described later.

In step 1503, the learning information selection unit 260 receives the sensor data of the sensor #s by the sensor reception unit 220.

In step 1504, the learning information selection unit 260 uses the sensor data analysis unit 230 to perform frequency analysis on the sensor data of the sensor #s. In step 1504, the learning information selection unit 260 uses the values set in the fields of the sampling frequency 1608, the sampling period 1609, and the FFT length 1610 registered in the GW learning information management table 1600 to perform frequency analysis of sensor data. To implement.

In Step 1505, the learning information selection unit 260 reads the median value 1603, the lower limit value 1604, the upper limit value 1605, and the shift speed 1606 of the sensor #s from the GW learning information management table 1600. Since median 1603, lower limit 1604, and upper limit 1605 are updated each time learning information selection unit 260 is executed, median 1603, lower limit 1604, and upper limit obtained when the process of FIG. 15 was executed last time. 1605 is read out. However, the information read from the GW learning information management table 1600 when the processing of FIG. 15 is executed for the first time is the learning information transmitted from the central facility server 1.

In step 1506, the learning information selection unit 260 cuts out the frequency analysis result in the frequency range specified by the lower limit value 1604 and the upper limit value 1605 from the frequency analysis result of the sensor data analyzed in step 1504, and further cut out. The median is identified from the frequency analysis results obtained. The frequency range specified by the lower limit value 1604 and the upper limit value 1605 of the GW learning information management table 1600 is referred to as a “sensing range”.

For example, as a result of the frequency analysis performed in step 1504, the waveform shown in FIG. 13 is obtained, and the lower limit value and the upper limit value read in step 1505 are 7.8 kHz and 11.9 kHz, respectively. In this case, the learning information selection unit 260 extracts only the waveform in the range of 7.8 kHz to 11.9 kHz from the waveforms shown in FIG. An example of the extracted waveform is shown in FIG. Further, the learning information selection unit 260 specifies a median value (frequency having the highest vibration intensity) from the extracted waveforms. In the example of FIG. 19, since the frequency with the largest vibration intensity is 9.9 kHz, the median is specified as 9.9 kHz.

In the following description, the time (current time) when the learning information selection unit 260 is executed this time is T1, and the time when the learning information selection unit 260 is executed last time is T0. Then, the median value specified by executing step 1506 this time is expressed as “fc — 1”, and the median value specified last time (time T0) is expressed as “fc — 0”. Note that the median value (fc_0) specified in the previous process is the median value 1603 of the GW learning information management table 1600 read out in step 1505.

In step 1507, the learning information selection unit 260 accumulates the sensor data waveform cut out in step 1506 in the sensor data accumulation unit 270. Conversely, the portion of the waveform that was not cut out in step 1506 may be discarded at this point.

In step 1507, whether or not the waveform of the sensor data cut out in step 1506 is stored in the sensor data storage unit 270 is based on information preset in the sensor data transmission 1611 of the GW learning information management table 1600. To be judged. If “None” is set in the sensor data transmission 1611, the learning information selection unit 260 does not store the waveform in the sensor data storage unit 270 in Step 1507, and “Yes” is set in the sensor data transmission 1611. In this case, the learning information selection unit 260 accumulates waveforms in the sensor data accumulation unit 270 in step 1507. The waveform of the sensor data stored in the sensor data storage unit 270 is transmitted to the central facility server 1 in step 1509.

As described above, the learning information selection unit 260 can suppress the increase in the data accumulation amount in the gateway 2 by accumulating data limited to a necessary frequency band by the processing of Step 1505 to Step 1507. .

In step 1508, the learning information selection unit 260 calculates the shift amount and shift speed of the frequency peak (median value). Hereinafter, the shift speed calculated here is expressed as “fv1now / s”.

In the present embodiment, the difference (fc_1−fc_0) between the median value (fc_1) specified by executing Step 1506 this time and the median value (fc_0) specified last time is determined as the “shift amount” of the frequency peak. Call. The frequency peak shift speed fv1now / s can be obtained by calculating (fc_1-fc_0) / (T1-T0). fc_0 is read in step 1505, and fc_1 is specified in step 1506. Therefore, the learning information selection unit 260 obtains the difference between the median specified in step 1506 and the median read from the GW learning information management table 1600 in step 1505, and divides it by (T1-T0). , Fv1now / s. After the shift speed fv1now / s is obtained, the learning information selection unit 260 registers the obtained shift speed in the current shift speed value 1607 of the GW learning information management table 1600.

In step 1509, the learning information selection unit 260 reads the lower limit value, upper limit value, and median value read from the GW learning information management table 1600 in step 1505, and the median value obtained by executing this time (time T 1) step 1506. Are used to find new lower and upper limits at time T1. Specifically, the learning information selection unit 260 obtains a new lower limit value and upper limit value by performing the following calculation.

The lower limit value and the upper limit value newly obtained in step 1509 are denoted as f1_1 and f2_1, respectively. Further, the lower limit value and the upper limit value obtained in the previous process are respectively expressed as f1_0 and f2_0. The learning information selection unit 260
f1_1 = f1_0 + (fc_1-fc_0)
f2_1 = f2_0 + (fc_1-fc_0)
F1_1 and f2_1 are calculated by performing the above calculation. That is, the learning information selection unit 260 adds the shift amount obtained this time to each of the lower limit value (f1_0) and the upper limit value (f2_0) stored in the GW learning information management table 1600, thereby obtaining the lower limit value and the upper limit value. Update the value.

The new lower limit value and upper limit value obtained in step 1509 and the new median value obtained in step 1506 are registered in the lower limit value 1604, upper limit value 1605, and median value 1603 of the GW learning information management table 1600, respectively. .

If “Yes” is stored in the sensor data transmission 1611, the learning information selection unit 260 transmits the sensor data accumulated in step 1507 to the central facility server 1 in step 1509. Conversely, when “none” is stored in the sensor data transmission 1611, the sensor data is not transmitted here.

In step 1510, the learning information selection unit 260 sets the absolute value of the shift speed fv1now / s obtained in step 1508 and the shift speed 1606 previously set in the GW learning information management table 1600 (referred to as “fv1 / s”). Compare absolute values. When the absolute value of fv1now / s is less than or equal to the absolute value of fv1 / s, the process proceeds to step 1503. When the absolute value of fv1now / s exceeds the absolute value of fv1 / s, the learning information selection unit 260 transmits information indicating that the shift speed is equal to or higher than the threshold to the central facility server 1 (step 1511). Transition to step 1501.

As described above, the shift speed 1606 set in the GW learning information management table 1600 is the shift speed of the waveform of the template selected by the central facility server 1, that is, a machine that has actually failed or a failure is likely to occur. This is the shift speed of the waveform obtained from the machine. If the absolute value of the shift speed obtained in step 1508 becomes larger than the absolute value of the shift speed of the waveform obtained from the machine in which the fault actually occurred (or the machine that is likely to fail), it means that the fault has occurred. It is estimated that Therefore, the learning information selection unit 260 according to the present embodiment performs the determination as in step 1510.

In addition, since the learning information selection unit 260 performs the process according to the above-described procedure, the information used when the gateway 2 detects the failure is only information included in the sensing range in the frequency analysis result of the sensor data. It is narrowed down. Thereby, increase of processing cost and accumulation cost can be controlled.

Also, the sensing range required for failure detection varies depending on the type of machine equipment to be diagnosed and installation conditions. As described with reference to FIG. 5, the data collection system according to the present embodiment determines the sensing range by comparing the sensor data collected from the sensor 5 with a plurality of templates. Therefore, information used for failure detection can be appropriately narrowed down.

A specific example of processing performed in steps 1505 to 1509 will be described with reference to FIGS. Hereinafter, an example in which the learning information selection unit 260 performs processing for the sensor # 1 will be described.

FIG. 16 shows the state of the GW learning information management table immediately after the learning information of the sensor # 1 sent from the central facility server 1 is set. The GW learning information management table shown in FIG. FIG. 18 shows a state immediately after the learning information selection unit 260 executes Steps 1505 to 1509 based on the GW learning information management table of FIG. 16, and the GW learning information management table shown in FIG. This represents a state immediately after the learning information selection unit 260 executes steps 1505 to 1509 based on the information management table. 16 to 18 show the contents of the GW learning information management table, but in order to avoid complicating the description, the reference numbers of the GW learning information management tables in FIGS. 1600, 1700, 1800. For the reference numbers of the columns, 1701 to 1711 are used for the GW learning information management table 1700 in FIG. 17, and 1801 to 1811 are used for the GW learning information management table 1800 in FIG.

For the sake of simplicity, the following description is based on the assumption that steps 1505 to 1509 are performed at 1 second intervals. In the following description, a case will be described in which the determination in step 1510 is always NO (fv1now / s is less than the shift speed specified by the central facility server 1).

The outline of processing performed by the learning information selection unit 260 after the learning information of the sensor # 1 sent from the central facility server 1 is set in the GW learning information management table 1600 is as follows.

The learning information of the sensor # 1 is transmitted from the central facility server 1 to the gateway 2, and step 1503 is executed when the learning operation state 1602 of the sensor # 1 becomes “operation”. The time at this time is T. In step 1503, the learning information selection unit 260 receives sensor data for one second and performs frequency analysis in step 1504 (that is, the time at the end of step 1504 is (T + 1)). At time (T + 1), steps 1505 to 1509 are performed, and the GW learning information management table 1600 is changed to the state shown in FIG.

Thereafter, the learning information selection unit 260 receives the sensor data for one second again and performs frequency analysis (steps 1503 and 1504). Then, at time (T + 2), when learning information selection section 260 executes steps 1505 to 1509, the state of GW learning information management table 1700 is changed to the state of FIG.

A change in the state of the GW learning information management table 1600 when Steps 1505 to 1509 are performed at time (T + 1) will be described with reference to FIGS. By executing Step 1505, the median value 1603 (10 kHz), lower limit value 1604 (8 kHz), upper limit value 1605 (12 kHz), shift speed (fv1 /) of the sensor # 1 are obtained from the GW learning information management table 1600 of FIG. s) is read (that is, the lower limit value, the upper limit value, and the shift speed sent from the central facility server 1 are read).

As a result, in step 1506 and step 1507, only the data in the range of 8 kHz to 12 kHz is extracted from the frequency analysis result obtained in step 1504 and stored in the sensor data storage unit 270. In the following description, as a result of step 1506 being executed, the specified median is assumed to be 9.9 kHz.

Here, when Step 1508 is executed, the shift speed (fv1now / s) is
It is calculated as (9.9-10) ÷ 1 = −0.1 kHz (= −100 Hz).

Subsequently, in step 1509, the learning information selection unit 260 calculates the lower limit value (f1_1) and the upper limit value (f2_1). The lower limit is
f1_1 = 8 + (9.9-10) = 7.9
While the upper limit is
f2_1 = 12 + (9.9-10) = 11.9
It becomes.

As a result, the median value, the lower limit value, and the upper limit value of sensor # 1 in the GW learning information management table 1600 are changed as shown in columns 1703 to 1705 in FIG.

Next, changes in the state of the GW learning information management table 1700 (FIG. 17) when Steps 1505 to 1509 are performed at time (T + 2) will be described with reference to FIGS. When step 1505 is executed, the median value 1603 (9.9 kHz), lower limit value 1604 (7.8 kHz), and upper limit value 1605 (11.9 kHz) of sensor # 1 are obtained from the GW learning information management table 1700 of FIG. The shift speed (fv1 / s) is read out. As a result, in step 1506 and step 1507, only the data in the range of 7.8 kHz to 11.9 kHz is extracted from the frequency analysis result obtained in step 1504 and stored in the sensor data storage unit 270. In the following description, the median value specified in step 1506 is 9.7 kHz.

Here, when Step 1508 is executed, the shift speed (fv1now / s) is
It is calculated as (9.7−9.9) ÷ 1 = −0.2 kHz (= −200 Hz).

Subsequently, in step 1509, the learning information selection unit 260 calculates the lower limit value (f1_1) and the upper limit value (f2_1). The lower limit is
f1_1 = 7.8 + (9.7-9.9) = 7.6
While the upper limit is
f2_1 = 11.9 + (9.7-9.9) = 11.7
It becomes.

As a result, the median value, lower limit value, and upper limit value of sensor # 1 in the GW learning information management table 1700 are changed as shown in columns 1803 to 1805 in FIG.

As described above, when the processing of FIG. 15 is executed for the first time, the learning information selection unit 260 uses the learning information (lower limit value and upper limit value) transmitted from the central facility server 1 from the waveform of the sensor data. ) Of the frequency range (sensing range) designated by () is extracted, and the abnormality of the machine 9 to be monitored is detected based on the extracted data. However, since the learning information selection unit 260 corrects the sensing range based on the analysis result of the sensor data (step 1509), the learning information selection unit 260 extracts the data of the corrected sensing range after the second time. become.

The vibration frequency of the monitored machine 9 gradually changes due to the influence of aging. For example, as in the example described above, the median tends to gradually decrease. When the learning information selection unit 260 extracts only the sensing range data transmitted from the central facility server 1 each time and determines abnormality detection from the extracted data, characteristic information to be detected (central Value) cannot be captured. For this reason, in the data collection system according to the present embodiment, information necessary for abnormality detection (central) is corrected by correcting the sensing range based on the analysis result of the sensor data (specifically, the amount of change in the median). Value) is not captured.

Finally, the overall flow of processing performed in the data collection system according to the present embodiment will be described. FIG. 20 is a diagram illustrating the processing flow of the learning information changing unit 120 of the central facility server 1 and the learning information selecting unit 260 of the gateway 2 and information exchanged between the two in the course of processing.

Since the details of the processing flow have been described as described above with reference to FIGS. 5 and 15, only the data exchange portion between the central facility server 1 and the gateway 2 will be described with reference to FIG.

The gateway 2 transmits the sensor data received from the sensor to the central facility server 1 (2301). This is a process corresponding to step 1501 in FIG.

On the other hand, as described with reference to FIG. 5, the central facility server 1 performs a learning process for determining a frequency range of sensor data used for abnormality detection. First, the central facility server 1 receives sensor data from the gateway 2 until the learning end time is reached, and compares it with a template (2302 to 2304). This is processing corresponding to step 501 to step 504 in FIG.

After the learning end time has elapsed, the central facility server 1 creates learning information (2305) and transmits the learning information to the gateway 2 (2306). This is processing corresponding to step 505 to step 509 in FIG.

When the learning information is transmitted to the gateway 2, the gateway 2 transits to the learning operation state, and performs a process of detecting an abnormality of the machine 9 using the sensor data and the learning information. As described with reference to FIG. 15 (steps 1503 to 1509), the gateway 2 analyzes the sensor data obtained from the sensor 5, obtains the median shift speed, and the obtained median shift speed is used as learning information. It is determined that the shift speed does not exceed the included shift speed (2311, 2312).

When it is determined that the shift speed is exceeded (2312: YES), the gateway 2 notifies the central facility server 1 of data including information on the shift speed excess (2313).

When the central facility server 1 receives a notification from the gateway 2 that the shift speed has exceeded the threshold, the central facility server 1 notifies the maintenance personnel etc. of the abnormality of the machine 9 via the screen of the central monitoring terminal 64 (2309). Specifically, the central facility server 1 notifies the maintenance staff or the like of the abnormality of the machine 9 by displaying, for example, an alert message indicating that the abnormality has occurred in the machine 9 on the screen.

1 Central facility server 2 Gateway 5 Sensor 70 WAN
300 Learning information change program 400 Learning information selection program

Claims

A data collection system that collects time-series data output from a sensor provided in a facility to be monitored,
The data collection system includes:
A learning information storage unit that stores a plurality of templates that are comparison data with the time series data;
A learning information change unit that determines an inspection range of the time series data by comparing the time series data with the plurality of templates,
A sensor data analysis unit that generates a frequency spectrum by performing frequency analysis on the time-series data;
Using the information related to the inspection range of the time series data, the frequency spectrum to be inspected is extracted from the frequency spectrum generated by the sensor data analysis unit, and the abnormality of the equipment using the frequency spectrum to be inspected A learning information selection unit for detection;
Having a data collection system.
The learning information storage unit has a median value at which the signal intensity is maximum in the frequency spectrum of the template, and a minimum frequency among the frequencies at which the signal intensity has a minimum value in the frequency spectrum of the template. A certain lower limit value, an upper limit value that is the maximum frequency among the frequencies at which the signal intensity in the frequency spectrum of the template takes a minimum value, is stored in association with each template,
The learning information changing unit compares the time-series data with the frequency spectrums of the plurality of templates, determines the template having the highest similarity with the frequency spectrum of the time-series data as a similar template,
Including the lower limit value and the upper limit value of the similar template in the information related to the examination range of the time-series data, and notifying the learning information selection unit;
The data collection system according to claim 1.
The learning information selection unit determines the frequency range specified by the lower limit value and the upper limit value of the similar template notified from the learning information change unit as an initial value of the inspection target frequency range,
a) When the sensor data analysis unit generates the frequency spectrum, extracting the frequency spectrum included in the inspection target frequency range;
b) identifying the median value of the extracted signal strength of the frequency spectrum;
c) calculating a shift amount and a shift speed of the median value using the median value specified in b) and the median value specified by the learning information selection unit last time;
d) updating the lower limit value and the upper limit value of the inspection target frequency range by adding the shift amount of the median value to the lower limit value and the upper limit value of the inspection target frequency range;
Repeatedly
The data collection system according to claim 2.
As a result of performing step c),
e) When the shift speed of the median value exceeds a predetermined threshold, an alert message that an abnormality has occurred in the facility is output.
The data collection system according to claim 3.
The data collection system includes:
A server having the learning information storage unit and the learning information change unit;
A gateway device having the sensor data analysis unit and the learning information selection unit;
The gateway device is connected to the server via a first network, and connected to the sensor via a second network;
A sensor receiver that receives the time-series data from the sensor; and a data transmitter that transmits the time-series data to the server.
The data collection system according to claim 3.
A gateway device provided between a sensor provided in a facility to be monitored and a server that collects time-series data output from the sensor,
A sensor receiver for receiving the time-series data from the sensor;
A data transmission unit for transmitting the time series data to the server;
A sensor data analysis unit that generates a frequency spectrum by performing frequency analysis on the time-series data;
A learning information receiving unit for receiving information on the examination range of the time-series data;
Using the information related to the inspection range of the time series data, the frequency spectrum to be inspected is extracted from the frequency spectrum generated by the sensor data analysis unit, and the abnormality of the equipment using the frequency spectrum to be inspected A learning information selection unit for detection;
A gateway device comprising:
Information on the inspection range of the time series data includes a lower limit value and an upper limit value of the frequency,
The learning information selection unit
The frequency range specified by the lower limit value and the upper limit value included in the information related to the inspection range of the time series data is determined as an initial value of the inspection target frequency range,
a) When the sensor data analysis unit generates the frequency spectrum, the step of extracting the frequency spectrum included in the inspection target frequency range from the generated frequency spectrum;
b) identifying the median value of the extracted signal strength of the frequency spectrum;
c) calculating a shift amount and a shift speed of the median value using the median value specified in b) and the median value specified by the learning information selection unit last time;
d) updating the lower limit value and the upper limit value of the inspection target frequency range by adding the shift amount of the median value to the lower limit value and the upper limit value of the inspection target frequency range;
Repeatedly
The gateway device according to claim 6.
As a result of performing step c),
e) If the median shift speed exceeds a predetermined threshold, notify the server that an abnormality has occurred in the equipment;
The gateway device according to claim 7.
A gateway device that receives time-series data output from a sensor provided in a facility to be monitored;
In a data collection system having a server connected to the gateway device through a network and storing a plurality of templates that are data for comparison with the time-series data,
A learning step in which the server determines a test range of the time-series data by comparing the time-series data and the plurality of templates;
The gateway device uses the information regarding the inspection range of the time series data determined in the learning step to extract a frequency spectrum to be inspected from the frequency spectrum of the time series data, and the extracted frequency spectrum An abnormality detection step of detecting an abnormality of the equipment using
The abnormality detection method characterized by performing.
The server is a median value that is the frequency at which the signal strength is maximum in the frequency spectrum of the template, a lower limit value that is the minimum frequency among the frequencies at which the signal strength takes a minimum value in the frequency spectrum of the template, An upper limit value that is the maximum frequency among the frequencies at which the signal intensity in the frequency spectrum of the template takes a minimum value is held in association with each of the templates,
The server is the learning step,
Comparing the time series data with the frequency spectrum of the plurality of templates, determining the template having the highest similarity to the frequency spectrum of the time series data as a similar template,
Including the lower limit value and the upper limit value of the similar template in the information related to the inspection range of the time-series data, and notifying the gateway device;
The abnormality detection method according to claim 9, wherein:
In the abnormality detection step, the gateway device is
After determining the frequency range specified by the lower limit value and the upper limit value of the similar template notified from the server as an initial value of the inspection target frequency range,
a) extracting the frequency spectrum included in the inspection target frequency range from the frequency spectrum of the time-series data;
b) identifying the median value of the extracted signal strength of the frequency spectrum;
c) calculating the shift amount and the shift speed of the median value using the median value identified in b) and the median value identified in the previous abnormality detection step;
d) updating the lower limit value and the upper limit value of the inspection target frequency range by adding the shift amount of the median value to the lower limit value and the upper limit value of the inspection target frequency range;
Repeatedly
The abnormality detection method according to claim 10, further comprising a step.
The abnormality detection step further includes
e) When the shift speed of the median value obtained in c) exceeds a predetermined threshold value, an alert message indicating that an abnormality has occurred in the equipment is output.
The abnormality detection method according to claim 11, further comprising a step.